计算机应用

• 人工智能与仿真 •    下一篇

基于残差注意力机制的点云配准算法

秦庭威1,赵鹏程2,秦品乐3,曾建朝4,柴锐1,黄永琦1   

  1. 1. 山西省太原市中北大学
    2. 赵鹏程
    3. 中北大学大数据学院
    4. 中北大学
  • 收稿日期:2021-07-22 修回日期:2021-10-13 发布日期:2021-11-01 出版日期:2021-11-01
  • 通讯作者: 秦庭威

Point cloud registration algorithm based on residual attention mechanism

  • Received:2021-07-22 Revised:2021-10-13 Online:2021-11-01 Published:2021-11-01

摘要: 针对传统点云配准算法精度低、鲁棒性差以及放疗前后癌症患者无法实现精确放疗的问题,提出一种基于残差注意力机制的点云配准算法(ADGCNNLK)。首先,在动态图深度卷积网络(DGCNN)中添加残差注意力机制来有效的利用点云的空间信息,减少信息损失;然后,利用添加残差注意力机制的·DGCNN提取点云特征,使用添加残差注意力机制的DGCNN网络在提取点云特征时不仅可以在保持点云置换不变性的同时捕捉点云的局部几何特征,也可以在语义上将信息聚合起来,从而提高配准效率;最后,将提取到的特征点映射到高维空间中使用经典的图像迭代配准算法(LK)进行配准,简称该算法为基于残差注意力机制的点云配准算法。实验结果表明,所提算法与迭代最近点算法(ICP)、全局优化的ICP算法(Go-ICP)和PointNetLK相比,在无噪、有噪的情况下配准效果均是最好,其中,在无噪情况下,与PointNetLK相比,旋转均方误差降低了74.75%,平移均方误差降低了47.50%;在有噪声的情况下,与PointNetLK相比,旋转均方误差降低了73.13%,平移均方误差降低了44.18%,说明所提算法与PointNetLK相比鲁棒性更强,并最终将其应用于放疗前后癌症患者人体点云模型的配准,辅助医生治疗,实现了精确放疗。

Abstract: Aiming at the problems of low accuracy and poor robustness of traditional point cloud registration algorithms and the inability of accurate radiotherapy for cancer patients before and after radiotherapy, Attention Dynamic Graph Convolutional Neural Network Lucas-Kanada (ADGCNNLK) was proposed. Firstly, residual attention mechanism was added to Dynamic Graph Convolutional Neural Network (DGCNN) to effectively utilize spatial information of point cloud and reduce information loss. Then, the DGCNN with the residual attention mechanism was used to extract point cloud features, The DGCNN network used the residual attention mechanism can not only capture the local geometric features of the point cloud while maintaining the invariance of the point cloud replacement when extracting the point cloud features, but also semantically aggregated the information to improve the registration efficiency ; Finally, the extracted feature points were mapped to a high-dimensional space ,and the classic image iterative registration algorithm Lucas-Kanada(LK) was used for registration, which was referred to as a point cloud registration algorithm based on residual attention mechanism. Experimental results show that compared with the Iterative Closest Point (ICP), Globally Optimal ICP (Go-ICP) and PointNetLK, the proposed algorithm has the best registration effect in the absence of noise and noise. Among them, In the absence of noise, compared with PointNetLK, the rotation mean square error is reduced by 74.75%, and the translation mean square error is reduced by 47.50%; in the case of noise, compared with PointNetLK, the rotation mean square error is reduced by 73.13%, The translational mean square error is reduced by 44.18%, indicating that the proposed algorithm is more robust than PointNetLK, and it is finally applied to the registration of human point cloud models of cancer patients before and after radiotherapy, assisting doctors in treatment, and realizing precise radiotherapy.

中图分类号: